is_train: True
ddp: False
stat: ts
root_path: ./
out_path: ./outputs/audio2pose/
project: s2g
e_path:  weights/AESKConv_240_100.bin
eval_model: motion_representation
e_name: VAESKConv
data_path: ./datasets/BEAT_SMPL/beat_v2.0.0/beat_english_v2.0.0/
test_ckpt: ./ckpt/new_540_shortcut.bin
data_path_1: ./datasets/hub/
pose_norm: True
cfg: configs/sc_model_config.yaml


mean_pose_path: ./mean_std/beatx_2_330_mean.npy
std_pose_path: ./mean_std/beatx_2_330_std.npy

mean_trans_path: ./mean_std/beatx_2_trans_mean.npy
std_trans_path: ./mean_std/beatx_2_trans_std.npy


vqvae_upper_path: ./ckpt/net_300000_upper.pth
vqvae_hands_path: ./ckpt/net_300000_hands.pth
vqvae_lower_path: ./ckpt/net_300000_lower.pth
vqvae_face_path: ./ckpt/net_300000_face.pth
vqvae_lower_trans_path: ./ckpt/net_300000_lower_trans.pth
use_trans: True

decay_epoch: 500

vqvae_squeeze_scale: 4 
vqvae_latent_scale: 5

vae_test_len: 32
vae_test_dim: 330
vae_test_stride: 20
vae_length: 240
vae_codebook_size: 256
vae_layer: 4
vae_grow: [1,1,2,1]
variational: False

# data config
training_speakers: [2] #[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30]
additional_data: False
cache_path: datasets/beat_cache/beat_smplx_en_emage_2_128/
dataset: beat_sep_lower
new_cache: False

# motion config
ori_joints: beat_smplx_joints
tar_joints: beat_smplx_full
pose_rep: smplxflame_30
pose_fps: 30
rot6d: True
pre_frames: 4
pose_dims: 330
pose_length: 128
stride: 20
test_length: 128
m_fix_pre: False


audio_rep: onset+amplitude
audio_sr: 16000
audio_fps: 16000
audio_norm: False
audio_f: 256
audio_raw: None


word_rep: textgrid
word_dims: 300
t_pre_encoder: fasttext


facial_rep: smplxflame_30
facial_dims: 100
facial_norm: False
facial_f: 0


id_rep: onehot
speaker_f: 0


batch_size: 128
lr_base: 2e-4
trainer: shortcut_rvqvae
 
rec_weight: 1
grad_norm: 0.99
epochs: 1000
test_period: 20